Prediction of self-heating in detergent powders for ...eprints.whiterose.ac.uk/145860/3/EuroDrying...
Transcript of Prediction of self-heating in detergent powders for ...eprints.whiterose.ac.uk/145860/3/EuroDrying...
-
This is a repository copy of Prediction of self-heating in detergent powders for application to spray dryer wall accumulations.
White Rose Research Online URL for this paper:http://eprints.whiterose.ac.uk/145860/
Version: Accepted Version
Article:
Maxfield, L, Martindejuan, L and Bayly, AE orcid.org/0000-0001-6354-9015 (2020) Prediction of self-heating in detergent powders for application to spray dryer wall accumulations. Drying Technology, 38 (3). pp. 395-409. ISSN 0737-3937
https://doi.org/10.1080/07373937.2019.1576727
© 2019 Taylor & Francis Group, LLC. This is an author produced version of a paper published in Drying Technology. Uploaded in accordance with the publisher's self-archivingpolicy.
[email protected]://eprints.whiterose.ac.uk/
Reuse
Items deposited in White Rose Research Online are protected by copyright, with all rights reserved unless indicated otherwise. They may be downloaded and/or printed for private study, or other acts as permitted by national copyright laws. The publisher or other rights holders may allow further reproduction and re-use of the full text version. This is indicated by the licence information on the White Rose Research Online record for the item.
Takedown
If you consider content in White Rose Research Online to be in breach of UK law, please notify us by emailing [email protected] including the URL of the record and the reason for the withdrawal request.
mailto:[email protected]://eprints.whiterose.ac.uk/
-
1
PヴWSキIデキラミ ラa SWノa-HW;デキミェ キミ DWデWヴェWミデ Pラ┘SWヴゲ aラヴ AヮヮノキI;デキラミゲ デラ Sヮヴ;┞ Dヴ┞Wヴ W;ノノ AII┌マ┌ノ;デキラミゲ
L. Maxfield1, L. Martin2, A.E. Bayly1
1 School of Chemical and Process Engineering, University of Leeds, Leeds, LS2 9JT, UK
2P&G Technical Centre Ltd., Whitley Road, Longbenton, Newcastle upon Tyne, NE12 9TS, UK
E-mail of the corresponding author: [email protected]
Abstract:
During the spray drying of detergent formulations, powder can accumulate on the inner
walls of the spray drying tower. Under certain conditions, when the accumulations are large
enough self-heating can occur, leading to unwanted charring and in severe cases to thermal
runaway. In this study, basket heating methods, namely the steady-state approach and
cross-point temperature method, were used to estimate the zero-order kinetics of the self-
heating reaction of a typical detergent formulation. It was found that the resulting kinetics of
these two methods were not in agreement, with this being attributed to the Cross-Point
Temperature methodげゲ ゲWミゲキデキ┗キデ┞ デラ Wヴヴラヴゲく The estimated kinetics were used in a 2D-axismmetric transient model of heat transfer within an oven heated basket of detergent
powder. Temperature-time profiles, and critical ambient temperatures, predicted by the
model agreed well with experimental measurements. The model was also used to explore
the value of using nth order kinetics estimated using a thermogravimetric measurement
technique. Theses kinetic showed no benefit versus the zero order kinetics and introduced
unnecessary complexity. Future work will see this model and the insights gained applied to
the problem of modelling self-heating in spray dryer wall accumulations.
Keywords: detergent powder, self-heating, thermal runaway, spray drying, cross-point
IミデヴラS┌Iデキラミ Self-heating is the process by which some materials can increase in temperature without the
application of an external energy source. An exothermic reaction within the material causes
the temperature to rise and the stability of these systems is a balance between the internal
heat generation and heat loss from the system boundaries. Many materials exhibit self-
heating behaviour including milk powder [1], coal [2], and biomass [3], and this behaviour
can cause problems in the processing and storage of these materials. During the spray drying
of detergent formulations, layers of powder can build-up on the hot internal walls of the
spray drying tower. If this build-up is of sufficient thickness and the local temperature high
enough, the powder in these layers can self-heat. This can cause the powder to char which
compromises the final product quality or. In the worst cases thermal runaway can occur,
whereby an uncontrollable exothermic reaction leads to a rapid and often dangerous
increase in temperature. Being able to understand this behaviour and predict the self-
heating of these materials aids in product quality assurance and the minimising of risks in
the process.
The self-heating behaviour of similar materials have previously been explored using a
number of methods. The long established method is the steady-state method based on
Frank-K;マWミWデゲニキキげゲ デheory of thermal explosions [4]. This is a basket heating method which was originally used to estimate self-heating kinetics for activated carbons [5]. The steady-
Formatted: Font: 10 pt
-
2
state method is an effective approach based on determining the critical ambient
temperatures, the temperature above which thermal runaway will occur, for different
basket sizes of material. This is a slow method requiring several experiments to yield a single
data point. Developed as a faster alternative, the cross-point temperature (CPT) method is
another basket heating method, first proposed by Chong et al. [1], and originally used to
estimate the self-heating kinetics of skimmed and whole milk powders. Since then, Sujanti et
al. [2] has used this approach to study coal, and has shown that the steady-state method and
the cross-point temperature method estimate similar kinetics. Malow and Krause [6] has
also compared these two methods, along with the differential scanning calorimetry (DSC)
based Ozawa-Flynn-Wall method to estimate the kinetics of coal dust, cork dust, riboflavin
and a detergent powder. This is also one of the few studies that addresses self-heating in
detergent powders. Little details were given with regards the detergent powder formulation,
but it was shown that the steady-state and CPT methods estimated similar kinetics, giving
activation energies of 102(±30)x103 J mol-1 and 112(±24)x103 J mol-1 respectively. These
errors are relatively large but this is possible due to only 4 and 5 points being determined for
the steady-state and CPT methods respectively.
This study draws on this previous work and seeks to determine the best means of estimating
the self-heating reaction kinetics of a typical detergent powder, such that the estimated
kinetics can be used in a newly developed numerical model and applied to the study of self-
heating in detergent powder accumulations. The reaction kinetics have been estimated
using three different methods. Two of these are the previously outlined steady-state and
cross-point temperature (CPT) methods, which are used to estimate zero-order reaction
kinetics. The third method is a thermogravimetric fitting method, adapted from that of Yang
et al. [7]. This method is used to fit nth order kinetics to mass loss data from
Thermogravimetric Analysis (TGA). The kinetics estimated using these methods are
compared and the advantages and disadvantages of each method explored.
These kinetics are applied in a newly developed 2D-axistmmetric model of transient heat
and mass transfer in a cylindrical basket of detergent powder. This model is adapted from
the models of Chen [8] and Chong and Chen [9], which were used to model self-heating in
milk powders and explore aspects of the cross-point temperature method. Heat and mass
transfer properties of the detergent powder system are measured and the comparison of
this model to experimental temperature profiles explored. The properties measured were
the self-heating reaction kinetics, the powder drying kinetics, and the effective heat transfer
coefficient of the oven. Both the numerical model and the two basket heating methods are
based on the same energy balance for a self-heating body, given by:
貢系椎 項劇項建 噺 倦椛態劇 髪 貢芸畦結貸帳眺脹 (1) Here, 貢 (kg m-3) is the bulk density of the powder, 系椎 (J kg-1 K-1) is the specific heat capacity, 倦 (W m-1 K-1) is the thermal conductivity, 芸 (J kg-1) is the heat of reaction, 畦 (s-1) is the pre-exponential factor of the zero-order Arrhenius reaction, 継 (J mol-1) is the activation energy, 迎 (J mol-1 K-1) the universal gas constant, 建 (s) the time, and 劇 (K) the temperature. In this energy balance the left hand term denotes the local rate of enthalpy change in the solid, the
first right hand side term denotes the conductive heat transfer in the solid, and the final
term denotes the heat generation of the zero-order exothermic reaction.
-
3
This model is then used to predict temperature-time profiles in these baskets of detergent
powder across a range of ambient temperatures. Critical ambient temperatures for baskets
of varying size are also predicted. These predictions are then compared with experimental
data. The influence of the specific heat capacity, and the differences in predicted
temperature profiles from using zero-order and nth order reaction models are also explored.
In future work, the knowledge acquired here will be applied to predicting self-heating in
spray dryer wall build-up.
M;デWヴキ;ノゲ ;ミS MWデエラSゲ
Detergent Powder Composition
This investigation explores the self-heating properties of a model detergent powder. The
detergent powder used in this investigation was produced by Procter and Gamble, consisting
of surfactant/s, polymer/s, and inorganic salts. This sample was designed as a model
compositions that is representative of the spray-dried powder present in a commercial
laundry detergent product. The surfactant used is linear alkylbenzene sulphonate, present at
levels of approximately 17%, and it is thought to cause the majority of the observed self-
heating behaviour in this formulation. A finer grade of inorganic salt was used to reduce the
composition variability that can be seen between particles and between different sizes of
particles [10]. The particle size distribution of the detergent powder was measured by
sieving. The median particle diameter, the diameter for which 50% of the particles are
smaller (経泰待), was measured as 328´m. This has implications in the spacing of the thermocouples used in the Cross-Point Temperature method experiments.
Steady-State Approach
The steady-state approach, sometimes referred to as the F-K method, initially developed by
Bowes and Cameron [5], is based on the steady-state dimensionless form of the energy
conservation equation (1). It is widely used for determining the zero-order reaction kinetics
of self-heating powders and is the basis of the British Standard BS EN 15118:2007. The
dimensionless form is obtained by defining the dimensionless temperature, 肯, the dimensionless length, 行, the dimensionless exponent, 糠, and the dimensionless parameter, 絞:
肯 噺 帳眺脹屯鉄 岫劇 伐 劇著岻 行 噺 掴挑 絞 噺 帳眺脹屯鉄 諦町凋挑鉄賃 結貸 曇馴畷屯 糠 噺 帳眺脹屯 (2) Where 捲 (m) is the position, 詣 (m) is a characteristic length, and 劇著 (K) the ambient temperature. The steady-state energy conservation equation can then be expressed as:
椛逐態肯 噺 伐絞結 提怠袋提底 (3) The Frank-Kamenetskii parameter, 絞, is a dimensionless term that encompasses all the quantities required to describe the problem of self-heating, with a value above the critical
value, 絞頂追, leading to thermal runaway. The value of 絞頂追 is a function of geometry, the dimensionless exponent 糠 in equation (2), and the Biot number, 稽件, used in defining the dimensionless boundary conditions of this problem. Rearranging 絞 in equation (2) gives:
-
4
ln 峭絞頂追劇著┸頂追態詣態 嶌 噺 ln 磐貢芸畦倦 継迎卑 伐 継迎劇著┸頂追 (4) The steady-state approach is based on this equation. A stainless steel equi-cylindrical
(cylinder of equal height and diameter) mesh basket was filled with detergent powder and
placed in a Memmert UF75 forced convection oven. A type K thermocouple was inserted
into the centre of this basket to measure the core temperature. The oven temperature is set
and an additional thermocouple was placed near to the basket to measure the true ambient
temperature close to the basket. The temperature is measured over a number of hours to
determine if the powder undergoes thermal runaway at the oven set ambient temperature.
The test is repeated at different ambient temperatures, in 0.5°C increments, to find the
critical ambient temperature, above which the basket of powder undergoes thermal
runaway, for the size of basket used. Repeating this for baskets of different sizes and plotting
according to equation (4) allows the zero-order activation energy and pre-exponential factor
for the self-heating reaction to be estimated. In this investigation three equi-cylindrical
baskets were used, with diameters of 50mm, 60mm, and 70mm. 絞頂追 has previously been calculated for common geometries under the conditions of 糠 噺 タ and 稽件 噺 タ [11]. In reality, 絞頂追 is influenced by the finite values of 糠 and 稽件, and can be found by numerically solving equation (3), in a similar way to Parks [12]. 絞頂追 was found for different geometries and a range of values of 糠 and 稽件 by finding the greatest value of 絞 for which a steady state solution exists. A 1D implicit finite difference model was used to
calculate 絞頂追 for the infinite slab, infinite cylinder, and sphere, whereas a 2D implicit finite difference model was used to calculate 絞頂追 for the equi-cylinder. This value of 絞頂追 as a function of 糠 and 稽件 for an equi-cylindrical basket is illustrated in Figure 1. Interpolating this data allows the 絞頂追 value to be found for the 糠 and 稽件 values used in this study as shown in Table 1. This value is then used when experimentally determining the kinetics of the
detergent using equation (4).
Figure 1. 絞頂追 as a function of 糠 and 稽件 for an equi-cylindrical basket.
-
5
Table 1. 絞頂追 values for common geometries at ideal conditions (left) and at the conditions used in this investigation (right).
詩 噺 タ┸ 刷餐 噺 タ 詩 噺 惣匝┻ 掃想┸ 刷餐 噺 挿┻ 宋匝捜 Infinite Slab 0.878 0.720
Infinite Cylinder 2 1.587
Sphere 3.32 2.621
Equi-Cylinder (height=diameter) 2.844 2.218
Cross-Point Temperature (Transient) Method
The Cross-Point Temperature method is a transient means of determining zero-order self-
heating reaction kinetics developed by Chong et al. [1]. As a basket of powder is heated in an
oven the peripheral regions of the basket reach the oven temperature first. Heat conducts
slowly into the centre of the basket and as self-heating begins the core temperature rises
above that of the periphery, such that at some time the basket centre and an offset point
are at the same temperature. At this time the conduction between these two points is
assumed to be zero. The temperature at which this occurs is known as the cross-point
temperature (CPT) and at this point equation (1) reduces to:
ln 磐穴劇穴建 卑鞭脹退脹頓鍋畷 噺 ln 峭芸畦系椎 嶌 伐 継迎劇寵牒脹 (5) In this method a basket of powder with a number of type K thermocouples embedded in the
powder is heated in a Memmert UF75 forced convection oven. Of these thermocouples one
is placed at the geometric centre, and at least one more thermocouple is radially offset from
this. For a set ambient temperature the basket is heated and the temperature is recorded at
the centre and offset point for the duration of the experiment. From this the cross-point
temperature and heating rate at basket centre, 穴劇【穴建, is noted. Repeating this for a range of ambient temperatures and plotting the cross-point temperature and 穴劇【穴建 values for each test in accordance with equation (5) allows the zero-order kinetics to be estimated.
Unlike the steady-state approach, every test performed using this method yields a data
point. In this investigation, two equi-cylindrical baskets of diameter 50mm and 60mm were
used with an array of three thermocouples spaced at distances of 0mm, 6mm, and 12mm
from the basket centre, and at basket half-height, as shown in Figure 2. These
thermocouples are connected to a Pico Technology USB TC-08 data logger and sampled at a
rate of 1 measurement per second. These baskets were heated at temperatures ranging
from 222°to 235°C.
-
6
Figure 2. Basket of detergent powder and thermocouple setup.
Thermogravimetric Method
A method adapted from the DTG (differential thermogravimetry) method of Yang et al. [7]
was used to fit nth order kinetics to the mass loss curves of detergent powder samples
measured using thermogravimetric analysis (TGA) under ramped heating conditions. A
sample of powder was sieved into a number of size ranges such that the reactivity of each
size range could be compared. A sample of detergent powder, approximately 11mg in mass,
was placed into an aluminium pan and loaded into the TA Instruments Discovery TGA. These
samples were heated at 5°C min-1 from 50 to 500°C in air, with data sampled at a rate of 1
measurement every 0.5 seconds. Mass loss due to initial water evaporation and mass loss
which occurs after the reaction were removed from the curves and the data normalised. To
this curve the following equation for conversion, 糠, was fitted.
穴糠穴建 噺 岫な 伐 糠岻津畦結貸 帳眺脹 (6) The curve fitting toolbox built into MATLAB was used to fit this equation to the experimental
data. This tool uses the method of least squares to fit to this data. The best fitting values for
the activation energy, 継, the pre-exponential factor, 畦, and the order of the reaction, 券, were determined for each TGA experiment.
N┌マWヴキI;ノ MラSWノ This system was modelled using a set of 2D-axisymmetric partial differential equations in the
radial direction, 堅, and axial direction, 権, describing the heat transfer and mass transport in a cylindrical basket of powder. The transient energy conservation equation for this system
across a basket of radius 迎 (m) and height 傑 (m) in 2D cylindrical coordinates, adapted from that of Chen [8] and Chong and Chen [9], is given by:
貢系椎 項劇項建 噺 倦 峭項態劇項堅態 髪 な堅 項劇項堅 髪 項態劇項権態 嶌 髪 貢芸畦結貸帳眺脹 髪 貢鎚鳥茎塚 項隙項建 (7) 貢鎚鳥 (kg m-3) is the solid density of the powder particles, 茎塚 (J kg-1) is the heat of vaporisation of the liquid water in the particles, and 隙 (kg kg-1) is the moisture content of the particles on a dry basis. This form of the energy conservation equation makes use of zero-order kinetics.
-
7
If the higher order kinetics determined using TGA are to be used then the reaction term is
;Sテ┌ゲデWS ;IIラヴSキミェノ┞ デラ ;IIラ┌ミデ aラヴ デエW ヴW;Iデキラミげゲ SWヮWミSWミI┞ ラミ デエW IラミIWミデヴ;tion of reactive component, 潔追頂 (kg m-3), its initial concentration, 潔追頂┸待 (kg m-3), and the reaction order, 券. The reaction term in equation (7) is replaced with 釆潔追頂┸待 磐 頂認迩頂認迩┸轍卑津 芸畦結貸 曇馴畷挽, and the depletion in reactive component with time is expressed as:
項潔追頂項建 噺 伐潔追頂┸待 峭 潔追頂潔追頂┸待嶌津 畦結貸 帳眺脹 (8)
The moisture in the system is modelled in two states, firstly as liquid moisture in the
particles and secondly as vapour in the voids between the particles. The transport of water
vapour through the voids in the basket of particles is described by:
項桁項建 噺 綱経塚銚椎 峭項態桁項堅態 髪 な堅 項桁項堅 髪 項態桁項権態 嶌 伐 貢鎚鳥 項隙項建 (9) 桁 (kg m-3) is the water vapour concentration, 綱 is the void fraction or porosity, and 経塚銚椎 (m2
s-1) is the diffusion coefficient of vapour through the powder voids. 隙 (kg kg-1), the moisture content on a dry basis, can be expressed as:
伐貢 穴隙穴建 噺 月陳┸沈津畦椎券椎岫迎茎鎚桁鎚銚痛 伐 桁岻 (10) Here, 月陳┸沈津 (m s-1) is the internal mass transfer coefficient, 畦椎 (m2) is the average particle surface area, 券椎 (m-3) is the number of particle per unit volume, 桁鎚銚痛 (kg m-3) is the saturated vapour concentration, and 迎茎鎚 is the surface relative humidity. Drying of the powder was modelled using the Reaction Engineering Approach (REA). The REA was first proposed by
Chen and Xie [13] and is a means of modelling drying kinetics by applying chemical reaction
engineering principles. This approach assumes that evaporation is a first order activation
process with an energy barrier to overcome, taking the form of an activation energy, while
condensation is a zero order process without such a barrier. The REA is a simple and robust
approach capable of describing drying behaviour with minimal experiments required to yield
the necessary model parameters. The relationship between the REA and other drying theory
is unclear, but for this self-heating model where drying is not of critical importance this
approach is sufficient. This approach has been used extensively to model the drying of
droplets but has also been used to model the drying of layer materials such as silica gel
particle layers [13], thin layers of pulped kiwifruit [14], and layers of mango and apple tissues
with thicknesses of a few centimetres [15]. In this model, the following Arrhenius equation is
used to express the relative humidity at the particle surface:
迎茎鎚 噺 ex� 釆伐 弘継塚迎劇鎚 挽 (11) つ継塚 is a correction factor in the apparent activation energy accounting for the added difficulty in drying at low moisture contents. When water covers the entire surface of the
solid this correction term reduces to zero, with relative humidity increasing to unity and
drying reducing to evaporation from a pure water surface. Drying is not a critical element of
this model as it does not impact on the observed self-heating behaviour in these powders,
and as such only preliminary testing has been carried out. Using an approach based on that
-
8
of Chen [16] this REA model was fit to sorption isotherm data, and it was found that for the
detergent powder used in this investigation, the correction factor can be expressed as:
つ継塚 噺 ぱ┻ぬは捲など胎 ex�岷伐なぬ┻はの岫隙待┻待苔苔態腿岻峅 (12) It was found that the choice of values for parameters such as the internal mass transfer
coefficient, 月陳┸沈津 , and the vapour diffusion coefficient, 経塚銚椎 , have as much of an influence on the predicted drying behaviour as the drying model and correction factor used. Further
work is required to properly understand the drying behaviour in these systems. Despite this,
the drying model has been included to allow the self-heating model to be more easily
compared with experimental results.
Symmetrical boundary conditions are used along the central axes of the cylinder. In the
radial direction these are expressed as:
穴劇穴堅鞭追退待 噺 ど 穴桁穴堅鞭追退待 噺 ど (13) Similar gradients in the axial direction apply at 権 噺 ど. At the exposed boundaries heat transfer is dependent on the external effective heat transfer coefficient, 月 (W m-2 K-1), and mass transfer is dependent on the mass transfer coefficient, 月陳 (m s-1), such that in the radial direction:
伐倦 穴劇穴堅鞭追退眺 噺 月岫劇】追退眺 伐 劇著岻 伐綱経塚銚椎 穴桁穴堅鞭追退眺 噺 月陳 磐桁綱 鞭追退眺 伐 桁著卑 (14) Again, similar gradients apply in the axial direction at 権 噺 傑. The heat transfer coefficient for the oven was determined by using the transient temperature measurement method used by
Carson et al. [17], where the average effective heat transfer coefficient is back calculated
from the transient temperature measured at the core of a black aluminium cylinder. For this
oven, the convective heat transfer coefficient was correlated as: 月頂墜津 蛤 伐な┻ひぬ捲など貸替劇著態 髪ど┻などに劇著 髪 なぬ┻ばな (W m-2 K-1). The effective heat transfer coefficient is calculated by summing this with the radiative component for the powder basket.
This model was made in gPROMS ModelBuilder and uses in-built solvers to solve this set of
2D model equations. The basket radius and half height were each discretised into 31 points.
Having solved this model with discretisations of 62 and 124 points, it was found that 31
points was sufficient to achieve grid independence. The differential equations in this model
were solved using central finite difference methods, and a variable time step and variable
order implicit Backward Differentiation Formulae (BDF) based solver. The time step in this
solver is varied so that the following is satisfied:
彪 な券鳥 布 磐 香沈欠 髪 堅】捲沈】卑態津匂沈退怠 判 な (15)
-
9
Here, 券鳥 is the number of differential variables, 香沈 キゲ デエW ゲラノ┗Wヴげゲ Wゲデキマ;デW aラヴ デエW ノラI;ノ Wヴヴラヴ in the ith variable, 捲沈 is the current value of the ith variable, 欠 is the absolute error tolerance, and 堅 is the relative error tolerance. This means that error, 香沈, in the variable 捲沈 is not allowed to exceed 欠 髪 堅】捲沈】 over a single time step, where the default value for 欠 and 堅 of 1x10-5 was used.
This model verified against analytical solutions for heat transfer in an infinite cylinder
documented by Carslaw and Jaeger [18]. This was done by using a large ratio of axial to
radial length such that this model can be assumed to be for an infinite cylinder. The heat of
reaction and drying terms were removed for this verification. The model solution agreed
with the analytical solution for the discretisation in space and time used. The heat
generation aspect is validated later by comparing model predicted and experimentally
measured critical ambient temperatures.
RWゲ┌ノデゲ ;ミS SキゲI┌ゲゲキラミ
Basket Heating Methods Results
The steady-state approach and cross-point temperature method were carried out as
outlined previously with the results shown in Figure 3 and 4. The estimated kinetics,
accompanying confidence intervals, and critical ambient temperatures for a 60mm equi-
cylinder as calculated from these kinetics are given in Table 2. The steady-state approach is
often seen as the standard method for determining the self-heating reaction kinetics for
powder systems and has been used extensively to determine such kinetics for powdered
materials such as coal, sawdust, and milk powder. Using this approach an activation energy
of 125.3x103 J mol-1, and a ln岫芸畦【倦岻 value of 34.63 were estimated, and these kinetics can be used in the equation for 絞 in equation (2) to predict critical ambient temperatures for simple geometries of any size. This simple approach is often used when dealing with the
problem of the storage of large quantities of self-heating materials.
-
10
Figure 3. Steady-state approach results using a 50mm, 60mm, and 70mm equi-cylindrical
baskets.
The cross-point tTemperature method was developed by Chong et al. [1] as an alternative to
the steady-state approach, with each test taking far less time and yielding a result. In this
investigation an array of three radially spaced thermocouples were used. These
thermocouples, as labelled in Figure 2, are 劇怠 at 0mm, 劇態 at 6mm, and 劇戴 at 12mm from the centre. By having an array of three thermocouple it is possible to evaluate three different
cross-point temperatures, as shown in equation (16).
系欠嫌結 な┺ 劇怠 噺 劇態┹ 系欠嫌結 に┺ 劇怠 噺 劇戴┹ 系欠嫌結 ぬ┺ 伐なの劇怠 髪 なは劇態 伐 劇戴はつ堅態 噺 ど (16) Case 1 and 2 come from a symmetrical three-point finite difference approximation for the
second order radial temperature derivative term in equation (1). Case 3 comes from a
symmetrical five point approximation for the second order radial temperature derivative
term. In a perfect system a five point stencil reduces the order of the error in approximating
the second order derivative term from an order of 頚岫つ堅態岻 to 頚岫つ堅替岻. It is worth noting that a similar three thermocouple setup has been used in previous investigations [19], but in each
case a three-point second order finite difference approximation was applied to the three
thermocouples. In doing this the cross-point was measured at the position of 劇態, whereas in this approach, using symmetry, case 3 measures the cross-point temperature at the position
of 劇怠 (i.e. the basket centre). The results for each case are shown in Figure 4, along with the best fit linear regression to
the data from which the reaction kinetics are determined. Experiments run at oven
temperatures below 222°C exhibited little self-heating, making the cross-point temperatures
difficult to measure at these temperatures and more sensitive to errors in thermocouple
readings.
Figure 4. Cross-point temperature (CPT) method results for three cases using a 50mm and
60mm equi-cylindrical baskets.
-
11
Table 2. Summary of estimated kinetics for the two basket heating methods and the
corresponding critical ambient temperature for a 60mm equi-cylindrical basket.
撮 (J mol-1) 残仔 釆晒冊暫 挽 残仔 峪晒冊察使 崋 60mm 参算司餐嗣 (K) Steady-State
Approach 125.3 (±1.0) x103 34.63 (±0.25) - 221.3
CPT Case 1 139.7 (±14.9) x103 - 28.21 (±3.51) 229.1
CPT Case 2 115.5 (±6.4) x103 - 22.88 (±1.53) 223.9
CPT Case 3 144.4 (±21.2) x103 - 29.24 (±4.99) 229.9
It can be seen that each case gives different results. The activation energies of case 1 and 3,
139.7x103 and 144.4x103 J mol-1 respectively, are very different to the activation energy
estimated using the steady-state approach, 125.3x103 J mol-1. The kinetics of case 2, with an
activation energy of 115.5x103 J mol-1 and ln 岫芸畦【系椎岻 value of 22.88, are much more similar to the steady-state approach kinetics. They also predict a much closer critical ambient
temperature (223.9°C) to that observed experimentally for a 60mm equi-cylindrical basket
(221.3°C). Case 1 and 3 over predict this with values of 229.1°C and 229.9°C respectively.
Other work [2] [6] has shown that the steady-state approach and CPT method yield
comparable results for coal, cork dust, riboflavin, and a detergent powder. The activation
energies for this detergent formulation are greater than those of Malow and Krause [6],
although the difference in formulation is unknown. The associated error for the steady-state
and case 2 results in particular are considerably smaller than those of Mallow and Krause.
It can also be seen in Figure 4 that the data points for case 2 are much less scattered than
the other two cases. This is echoed in the confidence intervals for case 2 which are
considerably smaller than those of case 1 and 3, with the confidence intervals of case 3
found to be particularly large. These experiments were simulated using the numerical model
and the cross-point temperatures determined for the same three cases. The results of this
analysis are shown in Figure ヵ ;ミS ゲ┌ェェWゲデ デエ;デ デエWヴW ゲエラ┌ノSミげデ HW ゲ┌Iエ ; ノ;ヴェe discrepancy between the results of these three cases. It does, however, show that a small difference
should exists between the cases. This difference is found to increase with an increase in
thermocouple spacing. The approximation associated with the cross-point temperature
assumption and the finite difference approximations used in equation (16) are thought to be
the reason for this difference.
-
12
Figure 5. Model simulated cross-point temperature method experiments simulating the
same three thermocouple pairing cases as measured experimentally.
Uncertainty Analysis
The errors in this approach can be attributed to a combination of the error in thermocouple
readings (±0.25°C) and error in thermocouple placement (±1mm). In the cross-point region,
the difference in temperatures across the three thermocouples is quite small, and thus any
error in the measured temperatures, be it due to error in thermocouple readings or errors in
placement, will have a significant impact on the measured cross-points. Looking across all
the 50mm basket experiments, an increase in oven temperature of 1°C corresponds to an
average increase in the case 2 cross-point temperature of 1.32°C. With such small
differences in cross-point temperatures it is evident that an error in thermocouple readings
of ±0.25°C will have a significant impact. Errors in thermocouple placement are difficult to
quantify. It is difficult to verify their position because the thermocouples are inserted into
the powder after the basket is filled, and thus the approximated error in placement of ±1mm
could have a significant impact on the measured cross-points.
These issues account for the errors in measurements, but do not explain the differences
observed between the results of the three cases. This difference may be because of the
small number of particle diameters which fit into the 6mm gap between thermocouples 劇怠 and 劇態. The median particle diameter is 328´m, such that as few as 19 particles of median diameter could fall within this space, which would heavily influence the case 1 results. This
would explain why the confidence intervals associated with the case 2 results are much
smaller than those of case 1. A finer grade of inorganic salt was used in this detergent
formulation to reduce composition variability and help alleviate this issue. Similar
experiments conducted using other detergent formulations, not shown here, without
reduced composition variability have had much more pronounced issues when using the
same thermocouple spacing of 6mm, and in some cases this variability made it impossible to
fit kinetics to the results.
-
13
It can been seen from the equation for case 3 in equation (16) that the thermocouple 劇態 has a much greater influence than 劇戴 on the five point finite difference approximation for the second order temperature derivative. This explains why the case 3 results are more similar
to the case 1 results, with case 3 also being affected by composition variability and the
problem of the small number of particle diameters that fit between 劇怠 and 劇態. Thermogravimetric Method Results
The best fit of the nth order conversion equation (6) to the normalised experimental
conversion curves are illustrated in Figure 6. Samples of different particle size ranges were
tested as mentioned previously. All samples reacted similarly showing no significant
difference in kinetics across the size ranges measured. This means that the effects of the size
distribution of the powder from a reactivity point of view can be considered negligible. A
representative fit is shown in Figure 6 and the kinetics given in Table 3. These kinetics were
used in the numerical model, making use of the reactive component dependent equation
(8). Very little self-heating behaviour was exhibited when using these kinetics at
temperatures where notable self-heating would be expected.
Figure 6. nth order reaction model fitting to the entire TGA measured conversion profile of
SWデWヴェWミデ ヮラ┘SWヴ ヮ;ヴデキIノWゲ ラa ゲキ┣W бヲヱヲ´マ ;ミS аンヰヰ´マ ;デ ; エW;デing rate of 5°C min-1.
In reality, multiple reactions may be occuring within this sample and fitting to the entire
thermal degredation, as has been done here, may not be a suitable approach. The oven
heated baskets of powder at sub-critical temperatures reached a maximum core
temperature of 541K, which is below the reaction onset temperature seen in the TGA
experiments. For this reason it is thought that the self-heating seen to occur in the baskets is
largely caused by the initial portion of the reaction observed using TGA. Because of this, the
fitting approach was applied only to the initial portion of the normalised conversion curve,
for ど 半 糠 半 ど┻ね, as can be seen in Figure 7, with the kinetics given in Table 3.
-
14
These kinetics are more similar to those found using the basket heating methods shown in
Table 2. Critical ambient temperatures for the nth order kinetics were not determined
because thermal runaway is difficult to pinpoint when using an nth order numerical model as
the reaction is always curtailed by reactive component depletion.
Figure 7. nth order reaction model fitting to the entire TGA measured conversion profile of
SWデWヴェWミデ ヮラ┘SWヴ ヮ;ヴデキIノWゲ ラa ゲキ┣W бヲヱヲ´マ ;ミS аンヰヰ´マ ;デ ; エW;デキミェ ヴ;デW ラa ヵェC マキミ -1.
Table 3. Summary of estimated kinetics using the Thermogravimetric Fitting Method.
撮 (J mol-1) 冊 (s-1) 仔 TGA Fitting (Full Curve) 168.9x103 3.14x1011 1.531
TGA Fitting (Initial) 136.2x103 4.10x108 1.01
Numerical Modelling of Self-Heating Basket Systems
The developed model predicts temperature, moisture content, and vapour concentration
profiles in the radial and axial directions of a quarter portion of a cylindrical basket. The
predicted temperature profiles are of most interest in this investigation, with the drying
behaviour included to allow for a better comparison with experimental data. The evolution
of these temperature profiles with time can be seen in Figure 8, showing the capabilities of
the model and illustrating how the core of the basket increases in temperature due to self-
heating in a typical basket heating experiment.
-
15
Figure 8. 2D model calculated temperature evolution in 50mm equi-cylindrical basket at and
ambient temperature of 502.5K as self-heating causes the core temperature to rise.
The internal mass transfer coefficient, 月陳┸沈津, and the vapour diffusion coefficient, 経塚銚椎 , were unknown and were estimated for this study. Drying is not of critical importance to this study,
and as such these values were adjusted to provide reasonable agreement with the drying
period of the experimentally measured temperature profiles. The model predicted core
temperature and how this compares to the experimental data can be seen in Figure 9. The
ambient temperature profile in the model is imported from the experimental data. The
increase in temperature of the powder above the oven temperature is of most interest as
this is an indication of the self-heating occurring. The kinetics determined using the steady-
state method were applied in this model and the good agreement seen in this region
suggests that these self-heating kinetics are correct.
Figure 9. Comparison of model predicted basket core temperature profile and experimental
data at an oven temperature of 502.5 K. Kinetics for this model were estimated using the
steady-state method.
In this study, the specific heat capacity, 系椎, of the powder was unknown, and a known value for similar detergent powders was used as a first estimate. The specific heat capacity does
not feature in the equation for the Frank-Kamentskii parameter, 絞, in equation (2) and therefore does not influence the critical ambient temperature of these self-heating systems,
-
16
merely the thermal response time. The effect of varying the specific heat capacity is shown
in Figure 10. This figure shows the model predicted core temperature profile of the basket
using varying values of 系椎 and how this compares to the experimentally measured temperature. The temperature profile using a 系椎 value of 1200 J kg-1 K-1 is seen to compare best with the experimental data, with lower values than this giving a faster thermal
response.
Figure 10. Comparison of model predicted basket core temperature profiles to experimental
data using different specific heat capacity values.
The comparison of the model predicted temperature profiles for a range of different
ambient temperatures is shown in Figure 11. This figure shows the comparison for the three
thermocouples at radial distances of 0mm, 6mm, and 12mm. These plots are for four
experiments run at oven controlled ambient temperatures from 498.5K to 504.4K. Again
good agreement is seen across these experiments, suggesting that this model is very capable
of predicting these temperature profiles.
-
17
Figure 11. Comparison of model predicted temperature profiles to experimental data from
an array of three thermocouples at radial distances of 0mm, 6mm, and 12mm, and at oven
temperatures from 498.5 K to 504.4 K. Kinetics for this model were estimated using the
steady-state method.
Two different sets of kinetics and their accompanying models were used in this investigation,
these being zero-order kinetics as estimated using the steady-state approach, and nth order
kinetics as estimated using the Thermogravimetric approach. The steady-state approach
kinetics are essentially fit to the experimentally determined critical ambient temperatures.
This means that thermal runaway as predicted by the model occurs at the same
temperatures as in the experiments. This can be seen in the plot of the basket core
temperature profile at a sub-critical ambient temperature in Figure 12 and at a super-critical
ambient temperature in Figure 13. This also serves as a validation of the heat generation
aspect of the model. Despite this, the same peak temperature is not reached because in
reality this is influenced by the depletion of reactive component which is not accounted for
in the zero-order kinetics model.
As mentioned previously, the kinetics from fitting to the entire TGA conversion curve were
used in the model with little self-heating exhibited. Because of this the kinetics from the
initial portion of the reaction were used instead, with this model yielding better results.
Despite this, the model predicted temperature profiles are still significantly different to the
experimental profiles as can be seen in Figure 12 and Figure 13. The system is seen to react
quickly, but as the reactive component depletes the reaction slows and the temperature falls
towards the oven temperature. In reality, the oxygen dependency of the reaction and the
slow diffusion of oxygen into the basket core may inhibit the rapid increase in temperature
generated by the reaction, as seen in this model. Additionally, this model is highly
dependent on the value used for the initial concentration of reactive component, 潔追頂┸待, and this is difficult to quantify. Only the initial portion of the reaction was accounted for when
fitting these kinetics, and if it is assumed that further reactions occur at higher temperatures
then characterising these reactions and including them in the model may help in replicating
-
18
the observed behaviour. The added complexity of the nth order model, and the difficulty in
getting the model to replicate the observed behaviour makes the zero-order model the
preferred choice for modelling these systems. When modelling self-heating in oven heated
baskets or in spray dryer wall accumulations such complexity is not required and the zero-
order models provide sufficient results.
Figure 12. Comparison of model temperature profiles using zero-order (estimated using the
steady-state method) and nth order (estimated using TGA fitting to the initial degredation
profile) kinetics to experimental data for a 50mm equi-cylindrical basket at a sub-critical
temperature of 500K.
-
19
Figure 13. Comparison of model temperature profiles using zero-order (estimated using the
steady-state method) and nth order (estimated using TGA fitting to the initial degradation
profile) kinetics to experimental data for a 50mm equi-cylindrical basket at a super-critical
temperature of 500.5K.
CラミIノ┌ゲキラミゲ This study sought to determine the best means of estimating the self-heating reaction
kinetics of detergent powders. The kinetics determined from the steady-state approach
allow a model fit to the experimentally observed critical ambient temperatures to be
produced, fitting very well with observed temperature profiles at the early stages of the
reaction. The method however is slow with each test taking between 4 and 8 hours, and a
number of tests required to obtain a single data point.
The Cross-Point Temperature method is a faster alternative, but this approach is more
susceptible to errors. The estimated kinetics using this approach varied depending on the
spacing of the thermocouples used. A relatively large spacing of 12mm helps to alleviate
issues of composition variability. The kinetics estimated using this spacing predicted thermal
runaway for a 60mm equi-cylindrical basket at 223.9°C, only 2.6°C greater than the
experimentally measured critical ambient temperature of 221.3°C. A smaller spacing of 6mm
produced more variability and kinetics that significantly over-predicted this temperature at
229.0°C. A slight discrepancy in estimated kinetics is expected, as shown using model
simulated experiments, but the large discrepancy observed experimentally could be due to a
combination of errors in thermocouple placement (±1mm), thermocouple readings
(±0.25°C), and the variability caused by the small number of particle diameters that fits in
the smaller thermocouple spacing of 6mm.
A numerical model was developed to allow predictions in temperature profiles and critical
ambient temperatures to be made. Having estimated a number of parameters, and using a
zero-order reaction model with the kinetics estimated using the steady-state method, the
predictions of this model were found to agree well with the experimentally measured
temperature profiles. This agreement was shown for baskets at a range of ambient
temperatures.
The nth order kinetics found by fitting to the entire TGA profile exhibited little self-heating
behaviour when used in the model, while fitting to the initial portion of the profile gave
kinetics which exhibit more of the expected behaviour. This model can be improved in a
number of ways but it is thought that this may merely an over-complication. The much simpler zero-order model compared considerably better to experimental data and such a
model is usually sufficient to describe self-heating in these systems.
Future work will see the approaches explored here, and the developed self-heating model
used to explore the problem of self-heating in spray dryer wall accumulations.
AIニミラ┘ノWSェマWミデゲ This work was supported by the UK Engineering and Physical Sciences Research Council
(EPSRC) Industrial CASE award and Procter and Gamble.
-
20
RWaWヴWミIWゲ [1] Chong, L.V., Shaw, I.R., Chen, X.D., Exothermic Reactivities of Skim and Whole Milk
Powder Measured using a Novel Procedure, Journal of Food Engineering, 30, (1996),
185-196
[2] Sujanti, W., Zhang, D.K., Chen, X.D., Low-Temperature Oxidation of Coal Studied Using Wire-Mesh Reactors with both Steady-State and Transient Methods,
Combustion and Flame, 117, (1999), 646-651
[3] Caballos, D. C. C., Hawboldt, K., Hellleur, R., Effect of production conditions on self-heating propensity of torrefied sawmill residues, Fuel, 160, (2015)m 227-237
[4] Frank-Kamenetskii, D.A., Diffusion and Heat Transfer in Chemical Kinetics, Plenum Press, New York (1969)
[5] Bowes, P.C., Cameron, A., Self-Heating and Ignition of Chemically Activated Carbon, Journal of Chemical Technology and Biotechnology, 21, (1971), 244-250
[6] Malow, M., Krause, U., The overall activation energy of the exothermic reactions of thermally unstable materials, Journal of Loss Prevention in the Process Industries, 17,
(2004), 51-58
[7] Yang, J., Miranda, R., Roy, C., Using the DTG curve fitting method to determine the apparent kinetic parameters of thermal decomposition of polymers, Polymer
Degradation and Stability, 73, (2001), 455-461
[8] Chen, X.D., Self-Heating Behaviour of Low Moisture Content Particles に Modelling the Basket-Heating of Solid Particles and some aspects of the Cross Over Behaviour using
Milk Powder as an Example, The ANZIAM Journal, 43, (1999), 165-181
[9] Chong, L.V., Chen, X.D., A mathematical model of the self-heating of spray-dried food powders containing fat, protein, sugar and moisture, Chemical Engineering Science,
54, (1999), 4165-4178
[10] Farshchi, A., Hassanpour, A., Bayly, A.E., The structure of spray-dried detergent powders. Manuscript in preparation
[11] Bowes, P.C., Self-Heating: Evaluating and Controlling the Hazards, Elsevier, Amsterdam (1984)
[12] Parks, J. R., Criticality Criteria for Various Configurations of a Self-Heating Chemical as Functions of Activation Energy and Temperature of Assembly, Journal of
Chemical Physics, 34, (1960), 46-50
[13] Chen, X.D., and Xie, G.Z., Fingerprints of the Drying Behaviour of Particulate or Thin Layer Food Materials Established Using a Reaction Engineering Model,
Transactions of the Institute of Chemical Engineers Part C: Food and Bio-Product
Processing, 75 C, (1997), 213-222
[14] Chen, X.D., The Basics of a Reaction Engineering Approach to Modeling Air-Drying of Small Droplets or Thin-Layer Materials, Drying Technology, 26, (2008), 627-
639
-
21
[15] Putranto, A., Chen, X.D., Webley, P.A., Modeling of Drying of Food Materials with Thickness of Several Centimeters by the Reaction Engineering Approach (REA),
Drying Technology, 29, (2011), 961-973
[16] Chen, X.D., A New Water Sorption Equilibrium Isotherm Model, Food Research International, 30, (1997), 755-759
[17] Carson, J.K., Willix, J., North, M.F., Measurement of heat transfer coefficients within convective ovens, Journal of Food Engineering, 72, (2006), 293-301
[18] Carslaw, H.S., Jaeger, J.C., Conduction of Heat in Solids, Oxford University Press, Oxford (1959)
[19] Chen, X.D., and Chong, L.V., Some Characteristics of Transient Self-Heating Inside and Exothermically Reactive Porous Solid Slab, Process Safety and
Environmental Protection, 73 B, (1995), 101-107